Introduction
Finding faces in an image can be quite a challenging issue. There are some commercial and Open Source frameworks available to developers in order to make the task of finding faces easier. A commercially available library that I can recommend is Neurotechnology VeryLook. Probably the most famous Open Source framework for face detection is Intel's OpenCV (Open Computer Vision) which is capable of a whole lot more than just detecting faces. As it is based on the C++ programming language, it is not easy to use in .NET environment. EmguCV solves this issue by providing a .NET interface to OpenCV functions. While these frameworks are powerful, they are not so easy to use.
In this article, I will explain the simplest way of detecting faces by using the Open-Source library Accord.net, which is built on top of the AForge imaging library that was mentioned in my post about using web camera with VB.NET.
Background
I am using face detection while developing solutions for biometric enrollment systems. Finding a face is an essential step in making the proper facial photo by international standards (ICAO). Only standardized images can be used in national documents, such are biometric passports, visas and national IDs.
Using the Code
The code sample is an extract of the attached project and in this article, I will focus only on the main functionality - finding a face with Accord.net library.
Dim detector As HaarObjectDetector
Dim cascade As New FaceHaarCascade
detector = New HaarObjectDetector(cascade, 30)
First of all, we need to declare class HaarObjectDetector. This class encapsulates algorithm for finding different objects in picture. Objects are described in so called cascades. Many of them are available in open-source projects (face, left eye, right eye, both eyes, nose, mouth, ...) Accord.net already incorporates cascade for finding face objects FaceHaarCascade, so you don't need to deal with cascade files.
detector.SearchMode = ObjectDetectorSearchMode.Average
detector.ScalingFactor = 1.5
detector.ScalingMode = ObjectDetectorScalingMode.GreaterToSmaller
detector.UseParallelProcessing = True
detector.Suppression = 3
Dim sw As Stopwatch = Stopwatch.StartNew
Dim faceObjects As Rectangle() = detector.ProcessFrame(PictureBox1.Image)
sw.Stop()
There are some parameters that have to be set for HaarObjectDetector class. Most important property is SearchMode which tells the detector which method to use while searching. In the example above, we use Average mode. With this mode, we mark object as face if it is found at least three times, what is set in Suppression property.
For best results, you need to play a little bit with the above properties. Most of the time, I only search for one face in image, so I usually set SearchMode to Single and I set ScailingMode to GreaterToSmaller, as the face in my case is usually the biggest object in the picture.
Dim g As Graphics = Graphics.FromImage(PictureBox1.Image)
For Each face In faceObjects
g.DrawRectangle(Pens.DeepSkyBlue, face)
Next
g.Dispose()
PictureBox1.Invalidate()
The last step is to draw a rectangle around the found face. The return structure from the detector is an array of Rectangles. With the for statement, we iterate through all of the found face rectangles and draw them on the Graphics of the source image.
Points of Interest
You can now easily combine web camera image (Fabrika LAB: WebCam Video) and the face detection explained here. Wow!
